Science & Tech · AI · 2 hrs ago
Inception unveils Mercury 3, claiming faster performance than rival AI models
California-based Inception has unveiled Mercury 3, its latest language model.
The company says it performs better than OpenAI's GPT-6 Luna and Anthropic's Claude Haiku 5.5 on several benchmarks, while running four to eight times faster.
Unlike traditional language models, which generate text one piece at a time, diffusion models generate and refine multiple pieces in parallel.
Mercury 3 can handle up to one million tokens of information in a single request, nearly four times the capacity of its predecessor.
Faster responses could help in interactive tools such as tutoring, live translation and software agents, where delays can build across multiple steps.
An early customer says it cut response times for student feedback to about one second, and developers can access Mercury 3 through several platforms; Inception says it has begun training its next model.
Inception has unveiled Mercury 3, a diffusion language model the company says outperforms GPT-6 Luna and Claude Haiku 5.5 on quality benchmarks while running four to eight times faster.
The company says Mercury 3 generates more than 1,000 tokens per second on widely available NVIDIA GPUs.
The model has a context window of 1 million tokens, up from 260,000 tokens in Mercury 2.5.
Inception says Mercury 3 is more than 40 times larger than Mercury 1 while maintaining comparable processing speeds.
The model is available through the Inception API, Baseten, OpenRouter and Vercel.
- Who
- Inception, a California-based AI research and product company, unveiled Mercury 3.
- What
- The company says its new diffusion language model outperforms GPT-6 Luna and Claude Haiku 5.5 on quality benchmarks while running faster.
- When
- October 19, 2026.
- Where
- Not stated.
- Why
- Inception says Mercury 3 shows diffusion language models can become more capable without sacrificing speed.
This story does not have two clearly opposing sides.
Inception introduced Mercury 1.
Inception unveiled Mercury 3.
- Model
- Mercury 3
- Inference speed
- More than 1,000 tokens per second
- Context window
- 1 million tokens
- Size compared with Mercury 1
- More than 40 times larger
- Availability
- Inception API, Baseten, OpenRouter and Vercel
What people are saying
Since we launched Mercury 1, the question we’ve heard most is whether diffusion can scale. Mercury 3 is over 40 times larger, significantly more capable, and still runs at over 1,000 tokens per second.
Today, we’re releasing Mercury 3. Mercury 3 is by far our most powerful model and the most intelligent diffusion LLM ever released. It outperforms GPT-6 Luna and Haiku 5.5 on quality while running at 1000+ tok/sec.
Posts on X
1/4 Today, we’re releasing Mercury 3. Mercury 3 is by far our most powerful model and the most intelligent diffusion LLM ever released. It outperforms GPT-6 Luna and Haiku 5.5 on quality while running at 1000+ tok/sec.